The evolution of natural language processing has shown how scaling up the amount of data and utilizing better neural network architectures has led to better prediction results compared to human predictions. The technique of masking has been used to improve models like BERT and Roberta.